1. Field of the Invention
The invention relates to an integrated circuit device for use in a computer system, more particularly to an integrated circuit device that has a core controller, a bus bridge, a graphical controller and a unified memory control unit built therein.
2. Description of the Related Art
Referring to FIG. 1, a conventional personal computer system 1 is shown to comprise a central processing unit (CPU) 10, a host bus 11 connected to the CPU 10, a core logic 12 connected to the host bus 11, a memory bus 13 connected to the core logic 12, a system memory 14 connected to the memory bus 13, an input/output (I/O) bus 15 connected to the core logic 12, at least one peripheral device 16 connected to the I/O bus 15, an Advanced Graphical Port (AGP) bus 17 connected to the core logic 12, a stand-alone video graphic accelerator (VGA) card 18 connected to the AGP bus 17, and a monitor 19 connected to the VGA card 18. The VGA card 18 includes a VGA chip 181, a local frame buffer 182 formed from dynamic memory, and a flash memory 183 for VGA BIOS.
Referring to FIG. 2, it has been proposed heretofore in another conventional personal computer system 2 to discard the stand-alone VGA card, and mount the VGA chip 281 and the local frame buffer 282 directly on the system board (not shown) to reduce costs and simplify manufacture of the system board.
Referring to FIG. 3, it has also been proposed heretofore in still another conventional personal computer system 3 to employ a unified memory architecture (UMA) in order to result in more cost savings by reducing the system board space and the components on the system board (not shown). As shown, the personal computer system 3 comprises a CPU 30, a host bus 31 connected to the CPU 30, a core logic 32 connected to the host bus 31, an I/O bus 35 connected to the core logic 32, at least one peripheral device 36 connected to the I/O bus 35, an AGP bus 37 connected to the core logic 32, a VGA chip 38 connected to the AGP bus 37, a shared system memory 34, and a wired-or memory bus 33 interconnecting the core logic 32, the VGA chip 38 and the shared system memory 34. Since the VGA chip 38 shares the system memory 34 with the core logic 32, the need to provide a dedicated local frame buffer for the VGA chip 38 is therefore obviated.
However, in order to enable the VGA chip 38 and the core logic 32 to share the system memory 34, some protocol must be introduced for the VGA chip 38 and the core logic 32 to follow. There is thus a need to incorporate additional pins and interface signals into the VGA chip 38 and the core logic 32 for protocol communication. Referring again to FIG. 3, the VGA chip 38 issues the MREQ# signal to inform the core logic 32 of its need to use the memory bus 33 for data transmission. After internal arbitration by the core logic 32, the core logic 32 releases the memory bus control to the VGA chip 38, and indicates the released state to the VGA chip 38 via the MGNT# signal. At this time, the memory access cycle of the VGA chip 38 can proceed as long as the MGNT# signal is driven to an active state by the core logic 32. The memory access cycle of the VGA chip 38 is terminated when the MGNT# signal becomes inactive, except in cases where the VGA chip 38 issues a high priority signal (not shown) to the core logic 32.
Furthermore, since there is a switching penalty whenever the memory bus control is switched from the core logic 32 to the VGA chip 38, and vice versa, the wired-or architecture of the memory bus 33 will downgrade the system performance. FIG. 4 shows detailed pin constructions of the memory control signals which are driven by the core logic 32 and the VGAchip 38 onto the memory bus 33 for data transmission. FIG. 5 shows the switching overhead that is incurred whenever the memory bus 33 is switched between the core logic 32 and the VGA chip 38. As illustrated, the time period T1 between activation of the MREQ# signal and activation of the MGNT# signal depends on whether or not the memory bus is idle and on the internal arbitration algorithm of the core logic 32. The length of the time period T2, i.e. the length of the MGNT# signal, depends on the VGA data transmission length and on the presence of a memory request from other master devices with a higher priority than the VGA chip 38. In time period T3, if there is a memory request from another master device with a higher priority, the core logic 32 will cease to assert the MGNT# signal to inform the VGA chip 38 to stop its data transmission by deactivating the MREQ# signal. Otherwise, the core logic 32 will only cease to assert the MGNT# signal after the VGA chip 38 has finished its data transmission and has inactivated the MREQ# signal.
Because the memory bus 33 has the wired-or architecture, when one of the core logic 32 and the VGA chip 38 assumes control of the memory bus 33, it becomes responsible for driving all control signals to the system memory 34 to ensure proper functioning of the latter. Whenever the control of the memory bus 33 is switched from one master to another, the original master of the memory bus 33 should drive all the control signals of the memory bus 33 to a high voltage level for at least one clock cycle, and subsequently float the memory bus 33 by deactivating the output (o/p) enable pins of the corresponding memory bus signals (see FIG. 4) to avoid bus contention. In this manner, at least four clock signals are wasted due to bus switching when the control of the memory bus 33 is switched from one master to another. Moreover, in order to meet timing requirements of the system memory 34, such as RAS# pre-charge time, both the core logic 32 and the VGA chip 38 must finish the RAS# pre-charge time before they switch the memory bus 33 and after they get control of the memory bus 33 in order to avoid compatibility issues in the event that the core logic 32 and the VGA chip 38 are made by different chip vendors.
It is also noted that the conventional personal computer system 3 involves overhead in translating data from one bus protocol to another bus protocol, i.e. the generation of data in the form of the destination bus protocol. Synchronization penalty is further incurred if the clock domain of the source bus is different from that of the destination bus.
Translating the host data destined for the VGA chip 38 in the conventional personal computer system 3 normally comprises three phases: the initiate phase, the translated phase, and the response phase. The initiate phase starts from the cycle request from a current host bus owner to the generation of an intermediate request. The translated phase starts from the intermediate request to the completion of the cycle on the destination bus. The response phase starts from the completion of the cycle by a response agent on the destination bus to the completion of the cycle on the originating bus. In most cases, the initiate phase on the originating bus and the response phase on the destination bus perform at different clock domains. In the conventional personal computer system 3 of FIG. 3, the initiate phase is performed on the 100 MHZ X86-like host bus 31, while the response phase is performed on the 66 MHz AGP bus 37.
Therefore, in the event of an incoming transaction to the AGP bus 37, synchronization must be performed before entering the translated phase owing to communication between the two different clock domains. FIG. 6 illustrates a synchronizer 39 for performing the aforesaid synchronizing function. The synchronizer 39 is separated into an input stage 391 and an output stage 392. The input signal at the input stage 391 is sampled using an input clock. The output signal of the input stage 391 serves as an input to the output stage 392, and is sampled using an output clock. The penalty of the synchronization process depends on the skew between the input and output clocks. As shown in FIG. 7, if the rising edge of the output clock is near the rising edge of the input clock, less penalty is paid for the synchronization process, as indicated by the output signal (A) in FIG. 7. Otherwise, if the rising edge of the output clock is far from the rising edge of the input clock, more penalty is paid for the synchronization process, as indicated by the output signal (B) in FIG. 7.
Upon entering the translated phase for an AGP bus transaction after the synchronization process, the VGA chip 38 accepts the transaction and drives the response. Before entering the response phase, the communicating signal must be synchronized owing to communication between the two different clock domains. This synchronization process is the same as that between the initiate and translated phases, the main difference residing in the switching of the definitions of the input and output clocks.